Autonomous labs need a “USB moment” — a common protocol so any AI agent can drive any instrument. It's an active 2025–26 research front.
The bottleneck in scaling self-driving labs isn't the AI — it's the plumbing. Every instrument speaks its own dialect, so today's autonomous systems are hand-integrated one at a time. A wave of 2025–26 work proposes shared agent-to-instrument protocols and lab operating systems so any agent can drive any device, the way USB standardised peripherals.
Get an AI to plan an experiment and it's impressive. Get it to actually run on your specific thermocycler, reader, and liquid handler — each with its own drivers, file formats, and quirks — and you're back to bespoke engineering. Reviews of the field flag this integration burden as the main barrier to scaling autonomous labs beyond a single showcase setup [1].
The emerging answer is standardisation. Proposals like agent-to-instrument protocols [2] and AI-native lab operating systems [3] aim to give agents a uniform way to discover, command, and read back from instruments — so adding a device is configuration, not a rebuild. (These are early, mostly preprint-stage efforts, but the direction is clear.)
Autonomy scales when adding an instrument is a plug-in, not a project.
Standardisation is why “works with the instruments you already own” matters: the value of physical AI compounds when it can operate your existing bench without a custom integration for every box. Perception plus plain language is one route to that — a new workflow is a description, not a driver-writing project. See how that works →
